Left atrial functional measurements’ utility in predicting long‐term risk of atrial fibrillation after isolated CABG

Atrial fibrillation (AF) is the most common cardiac arrhythmia following coronary artery bypass grafting (CABG). We hypothesized that measures of left atrial (LA) function would be useful in predicting AF in patients undergoing CABG.


INTRODUCTION
Coronary artery bypass grafting (CABG) is the most performed major cardiovascular surgery procedure in the US, constituting half of all major cardiovascular procedures. 1 The most common postoperative arrhythmia following CABG is atrial fibrillation (AF). 2,3 AF is also the most common cardiac arrhythmia overall with increased risk of stroke. 4 Hence, being able to identify patients at risk of AF is key to prescribing anticoagulants and preventing ischemic strokes. With correct screening, the risk of post-operative complications can be reduced as the treatment and monitoring of the patient can be individually tailored. Currently, congestive heart failure, hypertension, age >75 years, diabetes, and previous stroke/transient ischemic attack (CHADS 2 ) are used to determine the appropriate anticoagulant treatment for patients to prevent strokes. This scoring system is associated to the development of post-operative AF (POAF) and is also used to predict AF. 5 However, an improved model for predicting AF in patients with a CABG may decrease mortality due to a decrease in the number of patients experiencing stroke.
Patients are routinely examined before CABG with an echocardiogram to identify valvular disease and to evaluate the systolic function.
Currently, the most widely used echocardiographic measurement in the clinic for predicting AF is the maximal volume of the left atrium (LAVmax). 6 Several studies have found that functional measurements of the left atrium (LA), such as the LA emptying fraction (LAEF) and the LA minimum volume index (LAVmin), are useful in predicting AF. 7,8 Since it seems that the LA functional measures can play a role in predicting AF, we hypothesized that: (1) The LA functional measurements would be better predictors than LAVmax for AF following CABG.
Furthermore, we hypothesized that: (2) LA functional measurements would be significant predictors of AF even in patients with a normal LA size.

Population
In total, 782 patients undergoing isolated CABG at Gentofte Hospital Baseline data on medicine, comorbidities, laboratory and angiographic findings were recorded by reviewing the hospital charts.

Ethics
Retrospective studies in Denmark do not require approval by a scientific ethics committee but the study was approved by the Danish Health Authority (ID: 3-3013-339/1).

Endpoint
The endpoint was the development of any form of AF, which was obtained through diagnostic codes (ICD10: I48.9) from the Danish National Patient Registry.

Biochemical analysis
The patients had blood samples drawn when admitted to the hospital.
These samples were analyzed for hemoglobin, creatine kinase MB, Creactive protein and creatinine.

Echocardiography
Echocardiography was performed at a median of 15

Statistics
Statistical analysis was performed using STATA 14 (StataCorp LP, College Station, TX). Stratification was done by the outcome of AF.
Categorical variables are displayed as total numbers and percentages and were compared using 2 -test. Continuous variables with a Gaussian distribution are displayed as a mean ± SD and were compared using Student t test. Continuous variables showing non-Gaussian distribution are displayed as median with interquartile ranges (IQR) and were compared using Wilcoxon rank sum test.
Univariable Cox proportional hazards regression analyses were used for the three measurements of the LA: LAVmin, LAVmax, and LAEF. Multivariable Cox regression analyses were also applied to adjust for confounders and to obtain fitted hazard ratios. This was performed in two models: (1) Adjusting for potential confounders identified within the cohort: gender, age, heart rate and hemoglobin, (2) adjusting for CHADS 2 score.
The association between LA measurements and outcome were tested for interaction with gender and hypertension.
Subgroup analyses were performed in patients with normal LA size (LAVmax <34 mL/m 2 ), in patients with no previous history of myocardial infarction (MI) and preserved LVEF (>45%), and furthermore in patients with no history of MI, preserved LVEF, and normal LA size. Univariable as well multivariable Cox regressions were applied in the same way as for the entire population.
Cox proportional-hazard models were constructed for this subgroup based on incidence rate of AF stratified by low and high LAEF and low and high LAVmin.
Harrell's C-statistics were calculated from univariable Cox regression to estimate the predictive value of the LA measurements.
Competing risk regression was performed to account for all-cause death and hospitalization for heart failure as competing events. Multivariable adjustments were similar to the Cox regression models.
Inter-and intraobserver variability for LAVmax, LAVmin, and LAEF was performed in 20 randomly selected patients. Reproducibility is expressed by bias coefficients (mean difference ± standard deviation) and coefficients of variation (CV).
A p-value of <.05 was cut off point for significance in the analyses.

RESULTS
During a median follow up time of 3.7 years (IQR: 2.6;4.9), 52 patients (9%) developed AF. Baseline characteristics of the patients are displayed in Table 1. The mean age was 67 years, LVEF was 50%, 84% were male, 68% were hypertensive, and 26% were diabetic.
The patients who developed AF showed a trend towards being older (70 years vs. 67 years) and had a lower CCS class but were otherwise similar with respects to clinical characteristics compared to the group free of AF.
LAEF was the only echocardiographic measure that differed significantly between the outcome groups at baseline.

Predictive value of the atrial functional measurements
Uni-and multivariable Cox regressions are displayed in Table 2. LAEF was the only significant predictor in the univariable and multivariable model 1. This was consistent when further adjusting for LV mass and creatinine. However, when adjusting for CHADS 2 score, LAEF did not remain significantly associated with AF, but showed a trend towards such an association. No other LA measures were significantly associated with AF. No effect modification was found for hypertension nor gender with respects to the association between AF and any of the three LA measures (p for interaction >.05 for all).
In patients with a normal-sized LA (n = 531 with 49 events), both LAVmin and LAEF were significant predictors of AF in the univariable model ( In patients with normal size LA, those with high LAEF (>47%) were not statistically less likely to develop AF during follow-up ( Figure 1). However, we found a significant association between high LAVmin (>11 mL/m 2 ) and risk of AF in this subgroup, such that high LAVmin posed an increased risk of AF: HR = 1.95 (1.08-3.51), p for logrank = .02 ( Figure 2).
In patients with no history of MI and preserved LVEF (n = 366 with 32 events), no LA measure was associated with AF in univariable or multivariable Cox regression (supplementary Table S1). In the same subgroup, among those with normal LA size (n = 331 with 30 events), we observed that LAVmin was significantly associated with AF in unadjusted analysis (HR = 1.09 [1.01-1.18], p = .022, per 1 mL/m 2 increase), and this was consistent after multivariable adjustments. No other LA measure was associated with AF in this subgroup analysis (supplementary Table S1).

Competing risk assessment
In the overall population, no LA measure was significantly associated with AF in competing risk regression when accounting for all-cause death and heart failure hospitalization (supplementary Tables S2-S3).
However, in the subgroup with normal size LA, LAEF and LAVmin were significantly associated with AF in univariable regression. However, only LAVmin was significantly associated with AF in all multivariable regression models (supplementary Tables S2-S3).

Reproducibility
Intraobserver variability was lower than for interobserver analysis, as would be expected, but overall reproducibility was generally high and with overall similar reproducibility for all three measures. For LAVmax, TA B L E 1 Baseline characteristics.

DISCUSSION
The main finding from the present study is that no echocardiographic measurement independently predict AF after CABG. However, LAVmin and LAEF are independent predictors of outcome in patients with a normal-sized LA based on their odds ratio for AF in the univariable and multivariable Cox regression analysis-and both were better than the conventionally used LAVmax. These echocardiographic measurements may be useful in predicting AF in patients with normal sized LA who are at higher risk of AF. This may be because LAVmax is a measure of LA structure rather than LA function. In contrast both LAEF and LAVmin are more related to LA function, since LAEF is an indirect measurement of the atrial ability eject blood into the LV and a larger LAVmin is equivalent to a stiffer LA without the ability to contract in the diastole leaving a large residual volume, which is hypothesized to be a direct contributing factor to developing AF. 10,11 LAVmax has been shown to be a significant predictor of AF, 12 however, we found in the present CABG cohort that LAVmax was not associated with subsequent AF. Although LA dilation can lead to AF, LA remodeling occurs later on as a result of chronically elevated LA F I G U R E 2 Risk of AF according to LAVmin in patients with normal size LA. X-axis displays follow up time, Y-axis displays the probability of staying AF free. The plot is stratified by high LAVmin and low LAVmin, with the cutoff being 11 mL/m 2 . The solid blue line represents the likelihood of staying AF free for the low LAVmin subgroup and the light blue area is the 95% confidence interval. The solid red line represents the likelihood of staying AF free for the high LAVmin subgroup and the light red area is the 95% confidence interval. The graph displays a statistically significant lower probability of staying AF free for the subgroup with high LAVmin (p-value for log rank = .02).
pressure. This is also reflected in this study, as only a minor subset of patients had a dilated LA (n = 14). Accordingly, it is important to investigate tools that can identify patients at risk at an earlier point in time. 11 Measures of LA function may be valuable in this regard as also observed in several previous studies. 6,[13][14][15] This is further emphasized when considering that acquiring LAEF and LAVmin only requires one additional measurement and is therefore feasible in clinical practice.
LAVmin and LAEF are independent predictors of outcome in patients with a normal-sized LA, indicating that some valuable information can be gained by evaluating these aspects of LA function. This may be because LAVmax is a measure of LA remodeling and chronically elevated LA pressure and provides no insights as to LA function.
In contrast, both LAEF and LAVmin are more related to LA function, since LAEF is an indirect measurement of the atrium's ability to transfer blood into the LV and a larger LAVmin is equivalent to a stiffer LA without the ability to contract in the diastole leaving a large residual volume, which is hypothesized to be a direct contributing factor to developing AF. 10,11 LAEF is also associated with the LV filling pressure and in turn, LA pressure, which is known to cause dilation of the LA. 13 As functional measurements, LAEF is significantly associated with elevated left ventricular end diastolic pressure (LVEDP), which can be a precursor for elevated LA pressure. As the atrium must perform a greater amount of work (LAEF active ) due to higher LA afterload and less active contraction with limited oxygen, the myocardium will experience greater oxidative stress, which promotes fibrosis. The formation of fibrosis is a known factor in the pathogenesis of AF, as it increases the conduction heterogeneity of the myocardium. 13 Other echocardiographic measurements of both structure and function have previously proven significant in predicting AF in CABG patients, such as LA diameter, epicardial fat and LA expansion indexsolidifying the fact that a pre-operative echocardiography is an important measure to risk stratify patients for post-OP AF. [16][17][18] LAEF was a significant predictor of outcome in uni-and multivariable models in this subgroup of patients with normal-sized LA. LAEF has previously been proven to be a significant predictor of AF in patients with ischemic stroke and patients undergoing radiofrequency catheter ablation. 19,20 Unfortunately, the aforementioned studies did not present data on LAVmin.
LAVmin has been shown to result in the highest predictive performance, as determined from the C-statistic, of the functional measurements when added to the CHADS 2 score. 13 The CHADS 2 score is constructed from categorical variables, which makes it easier for clinicians to use, but this also simplifies the risk factors for developing AF.
The added value beyond clinical parameters may imply that we will be able to identify patients at risk of AF at an even earlier point and better prevent its associated complications.

Clinical perspective
AF is associated with increased mortality due to increased risk of cardioembolic stroke. 21,22 When also considering the increasing prevalence of AF, it is important to accurately predict the risk of AF in the individual patient. 23

Limitations
We do not have insight as to the monitoring process of the patients as the endpoint was drawn from patient registers. Also, patients initially excluded due to postoperative AF events could have developed clinical AF later, and we did not account for this in the present study.
As this was a retrospective study, we cannot exclude the possibility of residual confounding.
We did not measure the LA volume at the p-wave and can therefore not exclude that more detailed information on passive versus active LAEF could provide valuable information on the risk of AF.
The retrospective design is an inherent limitation to this study since the granularity of clinical details is limited. I.e. we did not have information on whether patients had a history of congestive heart failure and we did not have pro-BNP measurements available. Our findings may therefore be subject to residual confounding.
It should also be kept in mind that the CHADS 2 score was originally developed to predict stroke in AF patients and is therefore not optimized for AF prediction, but has nonetheless been proposed as a clinical prediction tool for AF. 24

CONCLUSION
No echocardiographic LA measurement was an independent predictor of AF for the entire population. However, for the subgroup with normal LA volume, LAVmin and LAEF were significant predictors of AF.
These findings should be investigated further in prospectively designed studies. The remaining authors have no disclosures to report.

DATA AVAILABILITY STATEMENT
In keeping with Danish legislation, the data underlying this study cannot be made publicly available because it contains sensitive patient data.